s2t 0.1.5__tar.gz → 0.1.6.post1.dev0__tar.gz

{s2t-0.1.5/src/s2t.egg-info → s2t-0.1.6.post1.dev0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: s2t
-Version: 0.1.5
+Version: 0.1.6.post1.dev0
 Summary: Speech to Text (s2t): Record audio, run Whisper, export formats, and copy transcript to clipboard.
 Author: Maintainers
 License-Expression: LicenseRef-Proprietary
@@ -55,6 +55,8 @@ System requirements (Linux)
   - Translate to English: `-t` (long: `--translate`). You may still provide `--lang` as an input-language hint if you want.
   - List available models and exit: `-L` (long: `--list-models`)
   - Recording format: `-f flac|wav|mp3` (long: `--recording-format`), default `flac`. MP3 requires ffmpeg; if absent, it falls back to FLAC with a warning.
+  - Auto-split on silence: `--silence-sec 1.0` (default `1.0`; `0` disables). When continuous silence ≥ this many seconds is detected, the current chunk is ended automatically.
+  - Minimum chunk length for auto-split: `--min-chunk-sec 5.0` (default `5.0`). Prevents very short chunks and avoids splitting early in a sentence.
   - Prompt mode (spoken prompt): `-p` (long: `--prompt`). Speak your prompt first, then press SPACE to use it as prompt and continue with your main content. If you press ENTER instead of SPACE, no prompt is used; the spoken audio is transcribed as normal payload and the session ends.
   - Keep chunk files: `--keep-chunks` — by default, per‑chunk audio and per‑chunk Whisper outputs are deleted after the final merge.
   - Open transcript for editing: `-e` (long: `--edit`) — opens the generated `.txt` in your shell editor (`$VISUAL`/`$EDITOR`).
@@ -69,6 +71,11 @@ Outputs are written into a timestamped folder under the chosen output directory
 - Final outputs: `recording.flac/.wav` (and `recording.mp3` if requested and ffmpeg available), plus `recording.txt/.srt/.vtt/.tsv/.json`
 - Clipboard mirrors the combined `.txt` with blank lines between chunks.
 
+Auto-splitting details
+- SPACE always splits immediately; ENTER finishes the recording.
+- With `--silence-sec > 0`, chunks end automatically after detected continuous silence of that many seconds.
+- Auto-split only triggers once the current chunk has at least `--min-chunk-sec` seconds and after speech has been detected (to ignore leading silence). A short internal cooldown avoids duplicate splits.
+
 ## Makefile (optional)
 - Setup venv + dev deps: `make setup`
 - Lint/format/test: `make lint`, `make format`, `make test`; combined gate: `make check`

{s2t-0.1.5 → s2t-0.1.6.post1.dev0}/README.md

@@ -27,6 +27,8 @@ System requirements (Linux)
   - Translate to English: `-t` (long: `--translate`). You may still provide `--lang` as an input-language hint if you want.
   - List available models and exit: `-L` (long: `--list-models`)
   - Recording format: `-f flac|wav|mp3` (long: `--recording-format`), default `flac`. MP3 requires ffmpeg; if absent, it falls back to FLAC with a warning.
+  - Auto-split on silence: `--silence-sec 1.0` (default `1.0`; `0` disables). When continuous silence ≥ this many seconds is detected, the current chunk is ended automatically.
+  - Minimum chunk length for auto-split: `--min-chunk-sec 5.0` (default `5.0`). Prevents very short chunks and avoids splitting early in a sentence.
   - Prompt mode (spoken prompt): `-p` (long: `--prompt`). Speak your prompt first, then press SPACE to use it as prompt and continue with your main content. If you press ENTER instead of SPACE, no prompt is used; the spoken audio is transcribed as normal payload and the session ends.
   - Keep chunk files: `--keep-chunks` — by default, per‑chunk audio and per‑chunk Whisper outputs are deleted after the final merge.
   - Open transcript for editing: `-e` (long: `--edit`) — opens the generated `.txt` in your shell editor (`$VISUAL`/`$EDITOR`).
@@ -41,6 +43,11 @@ Outputs are written into a timestamped folder under the chosen output directory
 - Final outputs: `recording.flac/.wav` (and `recording.mp3` if requested and ffmpeg available), plus `recording.txt/.srt/.vtt/.tsv/.json`
 - Clipboard mirrors the combined `.txt` with blank lines between chunks.
 
+Auto-splitting details
+- SPACE always splits immediately; ENTER finishes the recording.
+- With `--silence-sec > 0`, chunks end automatically after detected continuous silence of that many seconds.
+- Auto-split only triggers once the current chunk has at least `--min-chunk-sec` seconds and after speech has been detected (to ignore leading silence). A short internal cooldown avoids duplicate splits.
+
 ## Makefile (optional)
 - Setup venv + dev deps: `make setup`
 - Lint/format/test: `make lint`, `make format`, `make test`; combined gate: `make check`

{s2t-0.1.5 → s2t-0.1.6.post1.dev0}/src/s2t/cli.py

@@ -213,6 +213,8 @@ def run_session(opts: SessionOptions) -> int:
         verbose=opts.verbose,
         pause_after_first_chunk=opts.prompt,
         resume_event=prompt_resume_event,
+        silence_sec=opts.silence_sec,
+        min_chunk_sec=opts.min_chunk_sec,
     )
     t0 = time.perf_counter()
     chunk_paths, chunk_frames, chunk_offsets = rec.run(tx_q)
@@ -441,6 +443,18 @@ def main(argv: list[str] | None = None) -> int:
         default=0,
         help="Debounce window for SPACE (ms). If >0, ignores rapid successive space presses",
     )
+    parser.add_argument(
+        "--silence-sec",
+        type=float,
+        default=1.0,
+        help="Auto-split when continuous silence >= this many seconds (0 disables)",
+    )
+    parser.add_argument(
+        "--min-chunk-sec",
+        type=float,
+        default=5.0,
+        help="Minimum duration a chunk must reach before auto-split can trigger",
+    )
     parser.add_argument(
         "--native-segmentation",
         action="store_true",
@@ -496,6 +510,8 @@ def main(argv: list[str] | None = None) -> int:
             verbose=args.verbose,
             edit=args.edit,
             debounce_ms=getattr(args, "debounce_ms", 0),
+            silence_sec=getattr(args, "silence_sec", 1.0),
+            min_chunk_sec=getattr(args, "min_chunk_sec", 5.0),
             profile=args.profile,
             keep_chunks=getattr(args, "keep_chunks", False),
             prompt=getattr(args, "prompt", False),

{s2t-0.1.5 → s2t-0.1.6.post1.dev0}/src/s2t/config.py

@@ -18,6 +18,8 @@ class SessionOptions:
     verbose: bool
     edit: bool
     debounce_ms: int
+    silence_sec: float
+    min_chunk_sec: float
     profile: bool
     keep_chunks: bool
     prompt: bool

{s2t-0.1.5 → s2t-0.1.6.post1.dev0}/src/s2t/recorder.py

@@ -9,6 +9,8 @@ import time
 from pathlib import Path
 from typing import Any, Protocol, cast, runtime_checkable
 
+import numpy as np
+
 
 class Recorder:
     def __init__(
@@ -21,6 +23,8 @@ class Recorder:
         verbose: bool = False,
         pause_after_first_chunk: bool = False,
         resume_event: threading.Event | None = None,
+        silence_sec: float = 1.0,
+        min_chunk_sec: float = 5.0,
     ) -> None:
         self.session_dir = session_dir
         self.samplerate = samplerate
@@ -31,6 +35,9 @@ class Recorder:
         self.pause_after_first_chunk = pause_after_first_chunk
         self.resume_event = resume_event
         self._paused = False
+        # Auto-split config
+        self.silence_sec = max(0.0, float(silence_sec))
+        self.min_chunk_sec = max(0.0, float(min_chunk_sec))
 
     def run(
         self,
@@ -209,50 +216,65 @@ class Recorder:
             fh = sf.SoundFile(
                 str(cur_path), mode="w", samplerate=self.samplerate, channels=self.channels
             )
+            # State for auto-split based on silence
+            silent_frames_run = 0
+            seen_non_silent = False
+            last_split_time = 0.0
+            # Internal thresholds
+            threshold_rms = 0.015  # conservative RMS threshold for float32 [-1,1]
+            split_cooldown_sec = 0.2
+
+            def _do_split() -> None:
+                nonlocal fh, frames_written, cur_path, chunk_index, offset_seconds_total
+                fh.flush()
+                fh.close()
+                if frames_written > 0:
+                    dur = frames_written / float(self.samplerate)
+                    chunk_paths.append(cur_path)
+                    chunk_frames.append(frames_written)
+                    chunk_offsets.append(offset_seconds_total)
+                    offset_seconds_total += dur
+                    if self.verbose:
+                        print(
+                            f"Saved chunk: {cur_path.name} ({dur:.2f}s)",
+                            file=sys.stderr,
+                        )
+                    tx_queue.put((chunk_index, cur_path, frames_written, chunk_offsets[-1]))
+                else:
+                    try:
+                        cur_path.unlink(missing_ok=True)
+                    except Exception:
+                        pass
+                frames_written = 0
+                chunk_index += 1
+                if (
+                    self.pause_after_first_chunk
+                    and chunk_index == 2
+                    and self.resume_event is not None
+                ):
+                    self._paused = True
+                    self.resume_event.wait()
+                    self._paused = False
+                cur_path = self.session_dir / f"chunk_{chunk_index:04d}{self.ext}"
+                fh = sf.SoundFile(
+                    str(cur_path),
+                    mode="w",
+                    samplerate=self.samplerate,
+                    channels=self.channels,
+                )
+                # Reset silence tracking after a split
+                return
+
             while True:
                 # First, handle any pending control commands so SPACE/ENTER are never blocked by frames backlog.
                 try:
                     while True:
                         cmd = ctrl_q.get_nowait()
                         if cmd == "split":
-                            fh.flush()
-                            fh.close()
-                            if frames_written > 0:
-                                dur = frames_written / float(self.samplerate)
-                                chunk_paths.append(cur_path)
-                                chunk_frames.append(frames_written)
-                                chunk_offsets.append(offset_seconds_total)
-                                offset_seconds_total += dur
-                                if self.verbose:
-                                    print(
-                                        f"Saved chunk: {cur_path.name} ({dur:.2f}s)",
-                                        file=sys.stderr,
-                                    )
-                                tx_queue.put(
-                                    (chunk_index, cur_path, frames_written, chunk_offsets[-1])
-                                )
-                            else:
-                                try:
-                                    cur_path.unlink(missing_ok=True)
-                                except Exception:
-                                    pass
-                            frames_written = 0
-                            chunk_index += 1
-                            if (
-                                self.pause_after_first_chunk
-                                and chunk_index == 2
-                                and self.resume_event is not None
-                            ):
-                                self._paused = True
-                                self.resume_event.wait()
-                                self._paused = False
-                            cur_path = self.session_dir / f"chunk_{chunk_index:04d}{self.ext}"
-                            fh = sf.SoundFile(
-                                str(cur_path),
-                                mode="w",
-                                samplerate=self.samplerate,
-                                channels=self.channels,
-                            )
+                            _do_split()
+                            # Reset silence tracking on manual split
+                            silent_frames_run = 0
+                            seen_non_silent = False
                         elif cmd == "finish":
                             fh.flush()
                             fh.close()
@@ -289,6 +311,48 @@ class Recorder:
                     data = payload
                     fh.write(data)
                     frames_written += len(data)
+                    # Auto-split based on silence if enabled
+                    if self.silence_sec > 0.0:
+                        try:
+                            arr = np.asarray(data, dtype=np.float32)
+                            if arr.ndim == 2 and arr.shape[1] > 1:
+                                # average channels
+                                arr_mono = arr.mean(axis=1)
+                            else:
+                                arr_mono = arr.reshape(-1)
+                            # compute RMS
+                            rms = (
+                                float(np.sqrt(np.mean(np.square(arr_mono))))
+                                if arr_mono.size
+                                else 0.0
+                            )
+                        except Exception:
+                            rms = 0.0
+
+                        if rms < threshold_rms:
+                            silent_frames_run += len(arr_mono)
+                        else:
+                            silent_frames_run = 0
+                            seen_non_silent = True
+
+                        # Conditions to auto-split
+                        enough_silence = silent_frames_run >= int(
+                            self.samplerate * self.silence_sec
+                        )
+                        enough_length = frames_written >= int(self.samplerate * self.min_chunk_sec)
+                        cooldown_ok = (time.perf_counter() - last_split_time) >= split_cooldown_sec
+                        if enough_silence and enough_length and seen_non_silent and cooldown_ok:
+                            if self.verbose:
+                                print(
+                                    f"[auto] split (≥{self.silence_sec:.2f}s silence)",
+                                    file=sys.stderr,
+                                )
+                            last_split_time = time.perf_counter()
+                            # Queue a split for the next control phase
+                            ctrl_q.put("split")
+                            # Reset silence tracking now to avoid cascaded triggers
+                            silent_frames_run = 0
+                            seen_non_silent = False
             tx_queue.put((-1, Path(), 0, 0.0))
 
         def cb(indata: Any, frames: int, time_info: Any, status: Any) -> None:
@@ -302,7 +366,12 @@ class Recorder:
         key_t.start()
         writer_t.start()
 
-        print("Recording… Press SPACE to split, Enter to finish.")
+        msg = "Recording… Press SPACE to split, Enter to finish."
+        if self.silence_sec > 0.0:
+            msg += (
+                f" Auto-split on ≥{self.silence_sec:.2f}s silence (min {self.min_chunk_sec:.2f}s)."
+            )
+        print(msg)
         print("—" * 60)
         print("")
 

{s2t-0.1.5 → s2t-0.1.6.post1.dev0/src/s2t.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: s2t
-Version: 0.1.5
+Version: 0.1.6.post1.dev0
 Summary: Speech to Text (s2t): Record audio, run Whisper, export formats, and copy transcript to clipboard.
 Author: Maintainers
 License-Expression: LicenseRef-Proprietary
@@ -55,6 +55,8 @@ System requirements (Linux)
   - Translate to English: `-t` (long: `--translate`). You may still provide `--lang` as an input-language hint if you want.
   - List available models and exit: `-L` (long: `--list-models`)
   - Recording format: `-f flac|wav|mp3` (long: `--recording-format`), default `flac`. MP3 requires ffmpeg; if absent, it falls back to FLAC with a warning.
+  - Auto-split on silence: `--silence-sec 1.0` (default `1.0`; `0` disables). When continuous silence ≥ this many seconds is detected, the current chunk is ended automatically.
+  - Minimum chunk length for auto-split: `--min-chunk-sec 5.0` (default `5.0`). Prevents very short chunks and avoids splitting early in a sentence.
   - Prompt mode (spoken prompt): `-p` (long: `--prompt`). Speak your prompt first, then press SPACE to use it as prompt and continue with your main content. If you press ENTER instead of SPACE, no prompt is used; the spoken audio is transcribed as normal payload and the session ends.
   - Keep chunk files: `--keep-chunks` — by default, per‑chunk audio and per‑chunk Whisper outputs are deleted after the final merge.
   - Open transcript for editing: `-e` (long: `--edit`) — opens the generated `.txt` in your shell editor (`$VISUAL`/`$EDITOR`).
@@ -69,6 +71,11 @@ Outputs are written into a timestamped folder under the chosen output directory
 - Final outputs: `recording.flac/.wav` (and `recording.mp3` if requested and ffmpeg available), plus `recording.txt/.srt/.vtt/.tsv/.json`
 - Clipboard mirrors the combined `.txt` with blank lines between chunks.
 
+Auto-splitting details
+- SPACE always splits immediately; ENTER finishes the recording.
+- With `--silence-sec > 0`, chunks end automatically after detected continuous silence of that many seconds.
+- Auto-split only triggers once the current chunk has at least `--min-chunk-sec` seconds and after speech has been detected (to ignore leading silence). A short internal cooldown avoids duplicate splits.
+
 ## Makefile (optional)
 - Setup venv + dev deps: `make setup`
 - Lint/format/test: `make lint`, `make format`, `make test`; combined gate: `make check`